Granularity of OSError

[kj]

I've often come across the idea that good Python style deals with
potential errors using an EAFP ("easier to ask forgiveness than
permission") strategy rather than a LBYL ("look before you leap")
strategy.

For example, LBYL would look like this:

if os.path.isfile(some_file):
os.unlink(some_file)

In contrast, EAFP would look like this:

try:
os.unlink(some_file)
except OSError:
pass

But, as written, the EAFP code above is not satisfactory, because
there can be OSError's triggered for reasons other than the
non-existence of the regular file some_file.

What one needs is a finer granularity of exception, mapping to
exactly the error that one is guarding against.

Is there a standard approach to refining the granularity of exceptions
such as OSError? The only approach I can think of is to somehow
parse the error string (assuming is available) to determine whether
the exception is indeed of the specific kind we're trying to catch.
But this strikes me as grossly inelegant. Is there a better way?

(BTW, the problem is generic, because client code has no control
over the specificity of the exceptions raised by a library module.
If these exceptions turn out to be too broad, client code has to
somehow deal with this reality, at least in the short term.)

TIA!

kynn

 

[Sean DiZazzo]

I've often come across the idea that good Python style deals with
potential errors using an EAFP ("easier to ask forgiveness than
permission") strategy rather than a LBYL ("look before you leap")
strategy.

For example, LBYL would look like this:

if os.path.isfile(some_file):
    os.unlink(some_file)

In contrast, EAFP would look like this:

try:
    os.unlink(some_file)
except OSError:
    pass

But, as written, the EAFP code above is not satisfactory, because
there can be OSError's triggered for reasons other than the
non-existence of the regular file some_file.

What one needs is a finer granularity of exception, mapping to
exactly the error that one is guarding against.

Is there a standard approach to refining the granularity of exceptions
such as OSError?  The only approach I can think of is to somehow
parse the error string (assuming is available) to determine whether
the exception is indeed of the specific kind we're trying to catch.
But this strikes me as grossly inelegant.  Is there a better way?

(BTW, the problem is generic, because client code has no control
over the specificity of the exceptions raised by a library module.
If these exceptions turn out to be too broad, client code has to
somehow deal with this reality, at least in the short term.)

TIA!

kynn

You can access the exception object which gives you greater detail.

try:
os.unlink(some_file)
except OSError, e:
print e.errno
print e.strerror

if e.errno == 2:
pass
else:
raise


If it's the error you are looking for, handle it, or else raise.

~Sean

 

[Jeff McNeil]

You can access the exception object which gives you greater detail.

try:
    os.unlink(some_file)
except OSError, e:
    print e.errno
    print e.strerror

    if e.errno == 2:
        pass
    else:
        raise

If it's the error you are looking for, handle it, or else raise.

~Sean

I do this myself in a lot of places, almost exactly like this. It's
slightly clearer to use 'if e.errno == errno.ENOENT' in my opinion,
but, whatever.

The only place I've run into issues with this is when dealing with
socket programming across operating systems. Python on Windows exports
all of the WSA errors when dealing with sockets, but 'WSAEHOSTUNREACH'
is not the same as 'ENETUNREACH.' In cases like that, you've got to be
careful to check for all possible variations of the same error if you
care about supporting Windows.

Thanks,

Jeff
mcjeff.blogspot.com

 

[kj]

You can access the exception object which gives you greater detail.

try:
os.unlink(some_file)
except OSError, e:
print e.errno
print e.strerror

if e.errno =3D=3D 2:
pass
else:
raise

If it's the error you are looking for, handle it, or else raise.

Thanks, that's very handy.

kynn

 

[Ryan Kelly]

You can access the exception object which gives you greater detail.

I do this myself in a lot of places, almost exactly like this. It's
slightly clearer to use 'if e.errno == errno.ENOENT' in my opinion,
but, whatever.

In some cases it's also more correct. While ENOENT is always 2, some
error codes differ between windows and posix. In general it's better to
use the constants from the errno module.

Ryan


--
Ryan Kelly
http://www.rfk.id.au | This message is digitally signed. Please visit
(e-mail address removed) | http://www.rfk.id.au/ramblings/gpg/ for details


-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (GNU/Linux)

iEYEABECAAYFAkq0FPwACgkQfI5S64uP50q+hgCePblMyVIe0PiiwcQnQm24uDvw
UXUAn0COm5XYrupJG0p/IZEeDslV4lPr
=2h0s
-----END PGP SIGNATURE-----

 

[Sean DiZazzo]

The two version aren't equal. The first one suffers from a race
condition which may lead to a severe security issue. The file may be
gone or replaced by a different file in the time span between the check
and the call to unlink().

Christian

Thanks for pointing that out. I never thought of it before.

~Sean

 

[Grant Edwards]

The two version aren't equal. The first one suffers from a race
condition which may lead to a severe security issue. The file may be
gone or replaced by a different file in the time span between the check
and the call to unlink().

IOW, just be cause you look before you leap, it doesn't mean
you're not going to land on anybody and have to ask for
forgiveness afterwards.

Since you always have to handle the error case, there's not
much point in checking first unless the error case has bad
side-effects that you're trying to avoid. In this case,
attempting to unlink a non-existent file has no bad
side-effects, so there's no point in checking before the
unlink.

 

[MRAB]

IOW, just be cause you look before you leap, it doesn't mean
you're not going to land on anybody and have to ask for
forgiveness afterwards.

Since you always have to handle the error case, there's not
much point in checking first unless the error case has bad
side-effects that you're trying to avoid. In this case,
attempting to unlink a non-existent file has no bad
side-effects, so there's no point in checking before the
unlink.

It doesn't mean that LBYL is always a bad idea. If, for example, you're
going to copy a file, it's a good idea to check beforehand that there's
enough space available for the copy. There might be other process
changing the amount of freespace available, but it's still a reasonable
check to do.

 

[kj]

If, for example, you're
going to copy a file, it's a good idea to check beforehand that there's
enough space available for the copy.

How do you do that?

TIA,

kynn

 

[MRAB]

How do you do that?

There's os.statvfs(...), although that's Unix only.

The point is that it's sometimes a good idea to do a cheap check first
before attempting an operation that's 'expensive' even when it fails.

 

[Grant Edwards]

There's os.statvfs(...), although that's Unix only.

The point is that it's sometimes a good idea to do a cheap
check first before attempting an operation that's 'expensive'
even when it fails.

It's also important to note than under some circumstances, the
side-effects of that particular failure (filling up a disk
partition) can be unpleasant -- particularly if you do it as
root. Incoming Mail gets bounced/dropped, cron jobs fail, etc.

 

[ryles]

The point is that it's sometimes a good idea to do a cheap check first
before attempting an operation that's 'expensive' even when it fails.

Strongly agree. Furthermore, with LBYL it's often easier to give a
user clearer error messages for common usage errors, rather than
waiting for an exception in a much lower-level place where it's less
clear to them what the cause is.

 

[kj]

There's os.statvfs(...), although that's Unix only.

Thanks!

kynn